AMD A8-7600 Kaveri APU

The latest generation of AMD’s desktop CPU lineup features new Steamroller CPU and GCN GPU cores, helping it achieve improved power consumption. The A8-7600 has an additional energy efficiency trick up its sleeve: an adjustable 65W/45W TDP.

February 19, 2014 by Lawrence Lee

AMD A8-7600 FM2+ Processor
Street Price

AMD’s acquisition of ATI in 2006 paved the way for the Fusion microprocessor (now dubbed APU or accelerated processing unit), which combined both the CPU and GPU onto a single a chip. The move simplified chipset design while also making data transfer between the various components quicker, and cutting down manufacturing costs and power consumption. AMD had secondary motive though, to build toward a future where general purpose software would be written to utilize the hardware in the GPU portion of the chip in order to increase performance. Up until then, only games and CAD applications really took advantage of this extra processing capability. More often than not, most of those additional transistors sat idle most of the time — back then, even GPU-accelerated video playback was in its infancy.

The latest APU microarchitecture, codename Kaveri, is a showcase for Heterogeneous Systems Architecture (HSA). Kaveri was built to incorporate HSA, a system architecture designed to better mine the potential of APUs and System-on-Chips. For Kaveri, this meant bonding the CPU and GPU counterparts tighter than ever before so they could interact with fewer go-betweens. The two components have links that allow them share memory, caches, data queues, and other resources without stepping on each others toes. Work is scheduled and synchronized in a a more seamless fashion, taking the platform closer to true parallel computing. HSA can be found today in many ARM-based devices (tablets, phones, etc.) but AMD believes to belong on the desktop as well. Like AMD’s previous GPGPU (general purpose computing on GPUs) efforts, the problem is getting developers on board. For AMD’s vision to become reality, it needs mass adoption of this framework. To date, they’ve released four generations of this product line, with each boasting more of these features and new software titles capable of harnessing such capabilities but the library is still quite bare. It’s a long uphill battle with no end in sight.

The CPU-GPU landscape may eventually change, but until then we can still evaluate Kaveri on more traditional terms. Both the CPU and GPU portion of the new APU have received upgrades compared to its predecessor. Kaveri is the first to sport Steamroller CPU cores (the follow-up to Piledriver) which have two CPU modules with their own integer cores and a shared floating-point unit. Steamroller is manufactured using a new 28nm process which allows for higher transistor densities with lower switching speeds. This fabrication technology limits maximum frequency but it should more than make up for it in work per cycle and improved energy efficiency. The die size is almost identical to Richland but it sports an additional 1.1 billion transistors, with almost half of the die dedicated to the graphics processor which is now based on GCN (Graphics Core Next). It’s the same architecture found in the new Radeon Rx 200 line of discrete video cards, but new does not necessarily equate to better or faster. This version has but 6 or 8 compute units (depending on the model), not enough to propel it beyond entry-level performance.

Kaveri also brings PCI Express 3.0 to the table, though to really take advantage of that extra bandwidth multiple high-end cards are needed, which might be bottlenecked somewhat by the CPU power of a budget APU. In addition, AMD has added a TrueAudio DSP, dedicated hardware capable of offloading audio processing for various things like spatial effects and channel mixing. Like HSA, the benefit of TrueAudio will be solely determined by software developers — it’s a new piece of gear seeking adoption. AMD has a history of paying attention to sound, most notably for supporting HDMI audio and bitstreaming of Dolby TrueHD and DTS-MA audio on their graphics cards before their rival NVIDIA. HTPC enthusiasts were particularly appreciative of this and undoubtedly will also be grateful for Kaveri’s support for 4K resolutions, at 30 Hz over HDMI, and 60 Hz over DisplayPort.

The FM2+ socket.

The new design requires a new FM2+ socket which has two more pins than FM2, making Kaveri chips incompatible with older motherboards. You can drop a Trinity/Richland FM2 APU into a FM2+ board but this type of backwards compatibility is less desirable than the other way around. Upgrading the processor is obviously a much easier task and is typically more beneficial as well — this is doubly true if you’re currently running a Trinity/Richland system as the new motherboards don’t offer any added benefit as far as we can tell. AMD has been surprisingly hush on the new flagship A88X chipset, but from what we can gather from specifications of various A88X boards, nothing has changed from A85X.

The ASRock FM2A88X-ITX+.

For our review, we’ve been provided with a mini-ITX FM2+ motherboard made by ASRock, the FM2A88X-ITX+. This US$100 model is notable for having six SATA 6 Gbps ports, a headset amplifier chip, and a wireless 802.11n and Bluetooth 4.0 adapter (mini PCI-E).

The A8-7600.
Kaveri Desktop APU Comparison
A8-7600 (65W)
A8-7600 (45W)
CPU Cores
CPU Clock (Base/Turbo)
3.7 / 4.0 GHz
3.4 / 3.8 GHz
3.3 / 3.8 GHz
3.1 / 3.3 GHz
Total L2 Cache
Unlocked Multiplier
GPU Compute Units
Radeon Cores
GPU Clock
720 MHz
Price (USD)

AMD graciously sent us a single new APU for evaluation, the A8-7600, which occupies the bottom rung of AMD’s opening Kaveri salvo. The most interesting aspect of this chip is its adjustable TDP — it can be set to either 45W or 65W in the BIOS. Obviously users can achieve a similar effect by changing the CPU multiplier but you’d have to play around with it awhile to determine exactly how much power is being used. The TDP setting has the additional benefit of keeping the GPU within the same power envelope. All three launch models have 4MB of L2 cache and what AMD simply calls “R7” graphics with the premium A10-7850K sporting an extra two GPU compute units and 128 more cores. As usual, the K series has an unlocked multiplier for easier overclocking.

* EOL. Price of closet current comparative used.

When considering the cost of a system, the CPU is only part of the equation
as the price of motherboards varies greatly from platform to platform. In the
chart above, we added the price of the chips compared today to
those of an average compatible motherboard from Newegg.
The following criteria were used for the motherboards: retail versions, Asus, Intel (ASRock for AMD), Gigabyte, or MSI
branded, microATX or ATX form factor, SATA 6 Gbps and USB 3.0 support in some form, and only reasonably priced models were included. The average motherboard price turned out to be
US$98 for LGA1155, US$81 for FM2+ (Kaveri), and US$80 for FM2/FM2+ (Trinity/Richland).

At US$200, the A8-7600 configuration is one of the most affordable combinations compared. At the other end of the scale are a couple of Intel Sandy Bridge chips, the venerable i5-2500K and the low power i5-2400S. As these two chips are no longer being produced, we used the average street prices of their closest current comparatives.


Common CPU Test Configuration:

Common IGP Test Configuration:

AMD FM2 Platform:

  • AMD
    processor – 4.1 GHz, 32nm, 100W, integrated Radeon
    HD 8670D graphics
  • AMD A10-6700
    processor – 3.7 GHz, 32nm, 65W, integrated Radeon HD 8670D graphics
  • AMD
    processor – 3.8 GHz, 32nm, 100W, integrated Radeon
    HD 7660D graphics
  • AMD A10-5700
    processor – 3.4 GHz, 32nm, 65W, integrated Radeon HD 7660D graphics
  • AMD
    processor – 3.6 GHz, 32nm, 100W, integrated Radeon
    HD 7560D graphics
  • ASUS F2A85-M Pro motherboard
    – A85X chipset

AMD FM2+ Platform:

Intel LGA1155 Platform:

Discrete GPUs Compared: (using our GPU test system)

Measurement and Analysis Tools

Timed Benchmark Test Details

  • Photoshop: Image manipulation using a variety of filters, a derivation
    of Driver Heaven’s Photoshop
    Benchmark V3
    (test image resized to 4500×3499).
  • NOD32: In-depth virus scan of a folder containing 32 files of varying
    size with many RAR and ZIP archives.
  • WinRAR: Archive creation with a folder containing 68 files of varying
    size (less than 50MB).
  • iTunes: Conversion of an MP3 file to AAC.
  • TMPGEnc: Encoding a XVID AVI file with VC-1.
  • HandBrake: Encoding a XVID AVI file with H.264.

3D Performance Benchmarks

Video Test Suite

1080p | 24fps | ~22 mbps

H.264/MKV: A custom 1080p H.264 encoded clip inside an Matroska container.


1080p | 24fps | ~2.3 mbps

Flash 1080p: The Dark Knight Rises Official Trailer #3, a YouTube HD trailer in 1080p.

Testing Procedures

Our main test procedure is a series of both CPU (timed tests of real-world applications) and GPU-centric (gaming tests and synthetics) benchmarks. System power consumption is measured during the CPU tests (an average of the first 10~15 seconds) and in various states including idle, H.264 and Flash playback and full CPU and GPU load using Prime95/CPUBurn and FurMark.

Certain services and features like Superfetch and System Restore are disabled
to prevent them from affecting our results.
We also make note if energy saving features like Cool’n’Quiet and SpeedStep
do not function properly.

Estimating DC Power

The following power efficiency figures were obtained for the
Seasonic SS-400ET used in our test system:

Seasonic SS-400ET Test Results
DC Output (W)
AC Input (W)

This data is enough to give us a very good estimate of DC demand in our
test system. We extrapolate the DC power output from the measured AC power
input based on this data. We won’t go through the math; it’s easy enough
to figure out for yourself if you really want to.


Our first set of tests focuses on the integrated graphics. Each CPU/APU and motherboard combination was equipped with 4GB of RAM, a 500GB notebook hard drive and a Blu-ray drive.

IGP Energy Efficiency

With each generation, AMD’s desktop APUs have gained more energy efficiency. On light load, the Richland parts used slightly less power than Trinity, and Kaveri in turn, used slightly less than than Richland. The difference was only about 1~2W when idling and 3~4W during high definition Flash playback, but it’s nice to see continuing improvement.

Heavy load saw some noticeable decreases as well. Using the 65W TDP setting, the A8-7600 did not exceed 85W no matter what we threw at it. Like the A10-6700, while running Prime95, we weren’t able to push the load further by adding FurMark to load the GPU as well. TDP throttling kicked in, slowing the CPU cores down to 2.4 GHz, causing the power draw to actually drop by a modest amount. There was a third setting in the UEFI/BIOS to disable the TDP limit completely but it didn’t have any effect. The 45W setting limited power consumption to less than 70W.

According to CPU-Z, on full CPU load, the CPU core frequency ramped up to 3.7 GHz on the higher TDP setting, while the lower setting restricted clock speeds to 2.8~3.1 GHz.

IGP Performance – Synthetic

Note: Discrete GPUs were tested on our GPU testing platform which uses a
Core i3-2100, though CPU scaling shouldn’t be an issue given the relatively low level of GPU performance of the chips compared.

According to 3DMark11, the A8-7600 is head and shoulders faster than its predecessors, which doesn’t seem possible given the GPU hardware inside (on paper). The Heaven 3.0 benchmark had the 65W setting competing well with the A10-6700/6800K, while the 45W setting trailed well behind.

IGP Performance – Real World

Our real world gaming tests were conducted at two resolutions, 1366×768 (or 1280×800 if 1366×768 isn’t recognized as a valid resolution) and 1600×900, with differing levels of image quality. The results we’re reporting are for the highest resolution and detail level with which the product can deliver a reasonably good framerate (about 40 frames per second).

Note: Discrete GPUs were tested on our GPU testing platform which uses a
Core i3-2100, though CPU scaling shouldn’t be an issue given the relatively low level of GPU performance of the chips compared.

Unfortunately the encouraging synthetic results didn’t translate into better real world performance. In Aliens vs. Predator, it edged out all the other integrated solutions we’ve tested, but in Crysis it was dead last, while the rest of the test results were more mixed. It’s in the same class, more or less, as the 5000 and 6000 series. Playable framerates at 1600×900 resolution were achieved but keep in mind the games in our suite are older and not as demanding as cutting edge triple-A titles, so this is more of a best case scenario.

We arrived at our overall performance figures by giving each GPU a proportional
score with each game test having an equal weighting.
The scale has been adjusted so that the A8-7600 65W is the reference point with
a score of 100.

The A8-7600 65W falls directly into the middle of the pack, while at 45W, it’s at the bottom rung among AMD’s last three generations of desktop APUs. However, the overall spread between Kaveri, Richland, and Trinity, was only about 11 points. Given this tight range, the gaming performance of each APU shouldn’t be given much credence in any final buying decisions.


Our CPU testing is conducted with a discrete graphics card (a GeForce 9400 GT) to eliminate integrated graphics as a variable, most notably with regards to power consumption. It’s also necessary for fairly comparing CPUs that do not have an onboard graphics such as Bulldozer and Sandy Bridge Extreme models.

CPU Performance (Discrete Graphics)

Our CPU tests show the A8-7600, at 65W, is a middling contender, while at 45W, it was a few steps behind. The only interesting result was in WinRAR, with both the 65W and 45W setting struggling against the rest of the field. The new CPU cores seem to get more done per Hz but there is no substantial bump in performance. Energy efficiency has also improved, as the A8-7600 consistently used less power than its predecessors in each of our tests.

CPU Energy Efficiency (Discrete Graphics)

Power consumption on discrete graphics more or less mirrored what we saw when using the integrated graphics platform. Compared to Richland, there was a small boost in efficiency on light load and significant increase on heavy load. Each new family of APUs improved over the previous in this regard but it’s taken three years for AMD to finally match up to low power consumption of Sandy Bridge.

For users with balanced workloads, we’ve determined what we call the “average power consumption” which assumes the system is used half the time for light load activities (an average of idle and H.264 playback) and the remaining half for heavy load (an average of the power consumption used running our six benchmarks). We believe this is a very common usage pattern for an average PC — they are often left on for long periods of time, doing little to no work.

In this scenario, the A8-7600 at 45W edges out the i5-2400S by 3W, and at 65W, holds a 4W advantage over the i5-2500K.

For users with heavy workloads, the total power consumed while running our benchmark suite is of pertinent interest. The total power takes into account the energy efficiency of each CPU while running our benchmark tests as well as how quickly they complete each task. This simulates the power draw of a machine that is purely for doing work and shuts down when its job is finished.

In this metric, the A8-7600 gets a sizable bump over Richland but without better CPU performance, it still doesn’t come close to matching Intel’s offerings. Parity is still a long ways away.

CPU Performance Analysis

We arrived at our overall performance figures by giving each CPU a proportional
score in each real world benchmark with each test having an equal weighting.
The scale has been adjusted so that the A8-7600 65W is the reference point with
a score of 100.

The A8-7600 didn’t light the world on fire with its CPU performance, but at 65W and 45W, it did manage to come within striking distance of the A10-6700 and the A10-5700 respectively, despite having a significant disadvantage in clock speed. This bolds well for higher clocked parts like the A10-7850K (3.7 GHz). Give the A8-7600 a further 200 MHz bump and it should be competitive with the more expensive A10-6800K.

To calculate performance per dollar, we divided the overall performance score by the average platform costs we determined earlier and re-scaled it, again with the A8-7600 65W as our reference point.

AMD has been traditionally strong in this department (they’ve had to be) and the A8-7600 at 65W carries on this tradition, coming in second to only the A8-5600K. Even at 45W, there’s more value to be had than some of the lesser Richland and Trinity offerings.

To determine performance per watt, we took into account the average power consumption, again adjusted with the A8-7600 65W as the reference point.

Though arguably not faster than the previous Richland models we’ve tested, lower power consumption alone was enough to boost the A8-7600 ahead of AMD’s older APUs. The Intel chips remain king in this measurement, thanks to a combination of both excellent energy efficiency and performance.


Kaveri is another incremental improvement to AMD’s mainstream desktop lineup. Efficiency is the watchword, both in terms of CPU clock cycles and power consumption. The A8-7600, despite having a modest clock speed of 3.1 or 3.3 GHz (depend on the TDP setting) is competitive with last generation chips running 300~400 MHz faster. The newest APU iteration also manages to squeeze out a little extra energy efficiency, narrowing the gap with Intel’s contemporary parts. Despite the graphics processor undergoing an upgrade to the latest architecture, gaming performance is more or less equal to its predecessors.

Like Richland before it, Kaveri is a refinement of the AMD APU platform, bringing minor improvements. Everything is new and fresh, but when it comes down to it, the experience is more or less the same. Existing Richland and even Trinity users can hold off on upgrading, especially as a new motherboard is required. Unless you want 4K support, such a move would only provide marginal gains hardly worth the trouble and expense of changing sockets. For users looking for a new budget machine, however, we can certainly recommend an A8-7600 based system without any hesitation. It’s fast enough for most users, offers plenty of value, and if it’s not taxed strenuously, will shave a few dollars off your electric bill.

Kaveri is certainly preferable to the older generation in almost every regard but don’t expect much extra longevity. AMD’s recent track record with APU socket changes is not great; FM1 chips were launched in June of 2011, FM2 came out in October of 2012, and FM2+ was released in January of 2014. This 15~16 month window wouldn’t normally bother us but it does undermine the value of HSA. It’s essentially a form of future-proofing but there’s no guarantee developers will latch on to this model, and if/when they do, the FM2+ socket will likely be long gone. AMD may be building for tomorrow but users should really buy for today.

Our thanks to AMD for the A8-7600 and review sample used in this review.

* * *

Articles of Related Interest
AMD A10-6800K & A10-6700 Richland APUs
Intel Core i7-4770K Haswell Processor
AMD FX-8350 CPU: Piledriver Arrive
AMD A10-5700 APU: Trinity at 65W
AMD Trinity: A10-5800K & A8-5600K 2nd Gen APUs
Intel Core i7-3770 Ivy Bridge CPU

* * *

this article in the SPCR forums.

Leave a Comment

Your email address will not be published. Required fields are marked *